Classification of Quality Characteristics in Terms of Words using Review Rating Values

Understanding customer needs is crucial for improving products and services. Recent research has an approach that focuses on extracting valuable information from online review data, which is less expensive and easier to collect in large quantities. Previous studies have identified quality factors needing improvement and determined improvement priorities using the Kano model. However, these studies often assigned quality elements one by one to clusters of words limited by various conditions, leading to issues such as capturing only a small number of rough quality elements and missing elements that are difficult for researchers to assume but significantly affect customer satisfaction. To address these issues, this study proposes a method that treats all words in review sentences as candidate quality elements, capturing more detailed units of quality elements. Specifically, we combine Latent Dirichlet Allocation and logistic regression models to quantify the impact of each candidate quality factor on evaluation values and identify quality factors that need improvement. Candidate quality factors are captured both as individual words and as multi-word units, considering their affinities. To demonstrate the usefulness of the proposed method, we present the results of analyzing accommodation review data using these units.